Weld perception
Welding demands extreme precision. We developed vision algorithms that combine 2D images and 3D sensor data to localize weld seams in 3D space, correct their positions relative to the robot’s reference frame, and estimate weld profiles for highly precise robotic welding.
00:00
The positioner presents the joint while the robot travels along the pole.
Simulated segmentation on rendered camera images. Pixel coordinates are triangulated into 3D; the cross-section illustrates surface roughness, noisy sensor samples, and an estimated weld bead. Laser slices intersect the actual workpiece geometry.
Watch the pole-welding setup ↗